AI Pacing Debate Misses the Real Barrier: Enterprise Readiness
The debate over slowing frontier AI development overlooks a more fundamental problem: most companies are years away from being able to absorb even current-generation models, making the pacing argument largely irrelevant to near-term economic output.
The intensifying debate over whether to deliberately slow the development of frontier artificial intelligence models is asking the wrong question, according to a growing body of evidence from corporate boardrooms and economic research. The real constraint on AI's economic impact is not the speed of model development but the far slower pace at which businesses can restructure their data, workflows, and compliance systems to absorb the technology already available.
The argument, often framed as a race between the United States and China, has produced two camps. One side warns that any deliberate throttling of frontier model development amounts to unilateral disarmament. The other insists that pacing is the only responsible path for a technology whose own creators warn of catastrophic risk. Both positions, critics argue, rest on a shared misconception: the belief that every incremental leap in model performance translates immediately into macroeconomic output.
That assumption does not hold up against historical precedent. General-purpose technologies have consistently required decades to diffuse into measurable productivity gains. Electricity took roughly 75 years to lift productivity economy-wide. Computers required about 50 years. The internet and mobile devices demanded roughly 25. Artificial intelligence is following the same curve, even if the underlying models are advancing at an accelerated pace.
Corporate America, meanwhile, is already years behind the AI frontier. More than two-thirds of high-performing companies identify data as the primary barrier to implementing AI, a figure that has remained stubborn even as models have leaped forward. Only 7 percent describe their data as completely ready for AI. Fewer than a quarter have a data strategy at all, and 63 percent either lack AI-suitable data management or are unsure whether they have it.
Those numbers explain why the commercial fortunes of frontier labs will likely be decided by trust and adoption rather than raw capability. Many daily enterprise workflows require far simpler models, and precious few tasks at the average Fortune 500 company demand a frontier system at all. Older-generation chips, initially cast aside in the scramble for cutting-edge accelerators, are finding a second life as workhorses for the practical inference tasks that dominate enterprise demand.
One example came at a recent Yale CEO Caucus, where Growth Protocol founder and CEO Miro Dimitrov noted that deploying neuro-symbolic architectures allowed his enterprise reasoning platform to cut inference costs by roughly 80-fold in live client deployments, largely by shifting workloads off ultra-expensive GPUs and onto everyday enterprise CPUs.
McKinsey Senior Partner Asutosh Padhi has emphasized that technical availability is fundamentally different from economic transformation. When McKinsey surveyed the business community, it found that only 6 percent of companies reported a significant impact and modest earnings attribution. Companies are concentrating on high-reward, low-risk automation tasks that models one or two generations old can already solve.
The structural physics of enterprise architecture reinforce that caution. Fragmented data silos, legacy ERP systems, strict compliance regimes, and basic data hygiene make true economic absorption an inherently slow process. As one former Wall Street CEO put it, these systems will run in parallel with legacy systems for years to confirm they operate correctly and that no regulatory risk is unknowingly absorbed.
Corporate AI adoption is best understood in three phases, distinguished by how much work a company can responsibly hand over. The first phase, assistance, consists of off-the-shelf copilots that ride atop enterprise platforms, connecting data across existing applications and enabling employees to work faster with minimal re-architecting. Payback arrives quickly and risk stays modest, since a human still performs much of the work.
The second phase, orchestration, covers agentic workflows that demand real investment, including structuring proprietary data and connecting far-flung data lakes never meant to meet, with a human in the loop approving each consequential step. The third phase, autonomy, brings end-to-end agentic operations across seamlessly interconnected systems.
The implication for the pacing debate is straightforward. Slowing frontier development would cost the economy remarkably little in the near term, because enterprises need time simply to assimilate the capabilities already on the table. Racing ahead of alignment, by contrast, could cost far more. The argument, as currently framed, is not about how fast AI should advance. It is about how quickly institutions can change.
7
